demo/demo.py

import gradio as gr
from transformers import AutoTokenizer, AutoModel
from utils_MMD import extract_features  # Adjust the import path
from MMD_calculate import mmd_two_sample_baseline  # Adjust the import path

MINIMUM_TOKENS = 64
THRESHOLD = 0.5  # Threshold for classification

def count_tokens(text, tokenizer):
    """
    Counts the number of tokens in the text using the provided tokenizer.
    """
    return len(tokenizer(text).input_ids)

def run_test_power(model_name, real_text, generated_text, N=10):
    """
    Runs the test power calculation for provided real and generated texts.

    Args:
        model_name (str): Hugging Face model name.
        real_text (str): Example real text for comparison.
        generated_text (str): The input text to classify.
        N (int): Number of repetitions for MMD calculation.

    Returns:
        str: "Prediction: Human" or "Prediction: AI".
    """
    # Load tokenizer and model
    tokenizer = AutoTokenizer.from_pretrained(model_name)
    model = AutoModel.from_pretrained(model_name).cuda()
    model.eval()

    # Ensure minimum token length
    if count_tokens(real_text, tokenizer) < MINIMUM_TOKENS or count_tokens(generated_text, tokenizer) < MINIMUM_TOKENS:
        return "Too short length. Need a minimum of 64 tokens to calculate Test Power."

    # Extract features
    fea_real_ls = extract_features([real_text], tokenizer, model)
    fea_generated_ls = extract_features([generated_text], tokenizer, model)

    # Calculate test power list
    test_power_ls = mmd_two_sample_baseline(fea_real_ls, fea_generated_ls, N=N)

    # Compute the average test power value
    power_test_value = sum(test_power_ls) / len(test_power_ls)

    # Classify the text
    if power_test_value < THRESHOLD:
        return "Prediction: Human"
    else:
        return "Prediction: AI"

# CSS for custom styling
css = """
#header { text-align: center; font-size: 1.5em; margin-bottom: 20px; }
#output-text { font-weight: bold; font-size: 1.2em; }
"""

# Gradio App
with gr.Blocks(css=css) as app:
    with gr.Row():
        gr.HTML('<div id="header">Human or AI Text Detector</div>')
    with gr.Row():
        gr.Markdown(
            """
            [Paper](https://openreview.net/forum?id=z9j7wctoGV) | [Code](https://github.com/xLearn-AU/R-Detect) | [Contact](mailto:[email protected])
            """
        )
    with gr.Row():
        input_text = gr.Textbox(
            label="Input Text",
            placeholder="Enter the text to check",
            lines=8,
        )
    with gr.Row():
        model_name = gr.Dropdown(
            [
                "gpt2-medium",
                "gpt2-large",
                "t5-large",
                "t5-small",
                "roberta-base",
                "roberta-base-openai-detector",
                "falcon-rw-1b",
            ],
            label="Select Model",
            value="gpt2-medium",
        )
    with gr.Row():
        submit_button = gr.Button("Run Detection", variant="primary")
        clear_button = gr.Button("Clear", variant="secondary")
    with gr.Row():
        output = gr.Textbox(
            label="Prediction",
            placeholder="Prediction: Human or AI",
            elem_id="output-text",
        )
    with gr.Accordion("Disclaimer", open=False):
        gr.Markdown(
            """
            - **Disclaimer**: This tool is for demonstration purposes only. It is not a foolproof AI detector.
            - **Accuracy**: Results may vary based on input length and quality.
            """
        )
    with gr.Accordion("Citations", open=False):
        gr.Markdown(
            """
            ```
            @inproceedings{zhangs2024MMDMP,
                title={Detecting Machine-Generated Texts by Multi-Population Aware Optimization for Maximum Mean Discrepancy},
                author={Zhang, Shuhai and Song, Yiliao and Yang, Jiahao and Li, Yuanqing and Han, Bo and Tan, Mingkui},
                booktitle = {International Conference on Learning Representations (ICLR)},
                year={2024}
            }
            ```
            """
        )
    submit_button.click(
        run_test_power, inputs=[model_name, "The cat sat on the mat.", input_text], outputs=output
    )
    clear_button.click(lambda: ("", ""), inputs=[], outputs=[input_text, output])

app.launch()